Deliberative governance & anti-sycophancy
Reputation-weighted, commit-reveal aggregation that stops the AI from agreeing with itself — and records dissent instead of hiding it.
A known failure mode of multi-agent systems is sycophancy: agents converge on the first or loudest answer. Because LexCore's agents commit their findings independently and blind, it adds a disciplined reveal-and-aggregate step on top.
- 1Reputation — an EigenTrust-style computation weights agents by how much they corroborate each other over time, seeded by track record — not by how many findings they happen to file.
- 2Commit-reveal aggregation — findings about the same passage are clustered by position; the reputation-weighted majority becomes the consensus, and minority dissent is recorded, never dropped.
- 3Contest detection — when conflicting positions on one passage both carry real support, the passage is flagged contested and escalated to an arbiter, and failing that, to a Human Gate.
Trust
The recorded dissent is itself the anti-sycophancy audit trail. When two agents disagreed about a clause, you can see that they did — and how it was resolved.
Related technical documentation
Multi-agent debate (DyTopo)
A tiered bench of specialist agents produces and challenges findings in parallel rounds, ordered by a learning loop — so a claim survives scrutiny before you see it.
Human Gates & the triage worklist
Keeping a lawyer in the loop without drowning them: what escalates, and how the survivors are ranked by risk and materiality.
CitationGate: grounding every finding
The trust core. A finding with no verifiable quote is rejected; a quote that doesn't match the source is dropped. This is why LexCore's output is traceable.